@chrischall/gemini-mcp 0.9.0 → 1.0.1

package/.claude-plugin/marketplace.json

@@ -7,7 +7,7 @@
   },
   "metadata": {
     "description": "MCP server for Google Gemini image generation and editing — generate and edit images via the Gemini API",
-    "version": "0.9.0"
+    "version": "1.0.1"
   },
   "plugins": [
     {
@@ -15,7 +15,7 @@
       "displayName": "Gemini Images",
       "source": "./",
       "description": "Generate and edit images with Google Gemini image models (Nano Banana / Nano Banana Pro) via natural language prompts",
-      "version": "0.9.0",
+      "version": "1.0.1",
       "author": {
         "name": "Chris Hall"
       },

package/.claude-plugin/plugin.json

@@ -1,7 +1,7 @@
 {
   "name": "gemini-mcp",
   "displayName": "Gemini Images",
-  "version": "0.9.0",
+  "version": "1.0.1",
   "description": "MCP server for Google Gemini image generation and editing — generate and edit images via the Gemini API",
   "author": {
     "name": "Chris Hall",
@@ -19,6 +19,6 @@
     "google-gemini",
     "mcp"
   ],
-  "skills": "./SKILL.md",
+  "skills": "./skills/",
   "mcp": "./.mcp.json"
 }

package/README.md

@@ -4,7 +4,7 @@
 [![npm](https://img.shields.io/npm/v/@chrischall/gemini-mcp)](https://www.npmjs.com/package/@chrischall/gemini-mcp)
 [![license](https://img.shields.io/npm/l/@chrischall/gemini-mcp)](LICENSE)
 
-MCP server for Google Gemini image generation and editing. Exposes four tools to Claude over stdio: list available models, generate images from text prompts, edit or compose images with text instructions, and generate a consistent set of images from a master prompt. Images are written to disk by default (path returned) or returned inline as base64. Built on the Gemini v1beta API (`generativelanguage.googleapis.com`) using the Nano Banana / Nano Banana Pro model family.
+MCP server for Google Gemini media generation. Exposes eight tools to Claude over stdio: list available models, generate/edit/compose images, generate a consistent set of images from a master prompt, multi-turn image refinement (Interactions API), **video** generation (omni), **music** generation (Lyria), and an async result poll for long generations. Output is written to disk by default (path returned) or returned inline as base64. Built on the Gemini v1beta API (`generativelanguage.googleapis.com`) using the Nano Banana / Nano Banana Pro (images), omni (video), and Lyria (music) model families.
 
 Developed and maintained by AI (Claude Code).
 
@@ -42,10 +42,12 @@ progress), two guards make re-issuing safe and unnecessary:
 | Tool | Description |
 |------|-------------|
 | `gemini_list_models` | List available Gemini image models and the current default |
-| `gemini_generate_image` | Generate image(s) from a text prompt |
-| `gemini_edit_image` | One-off edits or multi-image composition with a text instruction (for a series of edits, use `gemini_interact`) |
-| `gemini_generate_set` | Generate a master image plus N consistent images referencing it |
+| `gemini_image_generate` | Generate image(s) from a text prompt |
+| `gemini_image_edit` | One-off edits or multi-image composition with a text instruction (for a series of edits, use `gemini_interact`) |
+| `gemini_image_set` | Generate a master image plus N consistent images referencing it |
 | `gemini_interact` | Preferred tool for iterative refinement: multi-turn generation/editing via the Interactions API — chain the returned `interaction_id` via `previous_interaction_id` (or `continue_last: true`) |
+| `gemini_video_generate` | Generate a short video (text→video, image→video, or `edit`) via the Gemini omni model (preview); written to disk as MP4 |
+| `gemini_music_generate` | Generate music from a text prompt via a Lyria model — `lyria-3-clip-preview` (~30s, default) or `lyria-3-pro-preview` (longer, WAV-capable); written to disk as MP3/WAV (preview) |
 | `gemini_get_result` | Fetch an async generation started with `async: true` by its `job_id` (status `running` → `done` result). Lets a long generation outlive a host's `tools/call` timeout |
 
 ## Quick Start
@@ -64,4 +66,4 @@ progress), two guards make re-issuing safe and unnecessary:
 }
 ```
 
-See [SKILL.md](./SKILL.md) for full usage documentation.
+See [SKILL.md](./skills/gemini-mcp/SKILL.md) for full usage documentation.

package/dist/bundle.js

@@ -32606,7 +32606,7 @@ var init_clipboard = __esm({
 init_dist();
 
 // src/version.ts
-var VERSION = "0.9.0";
+var VERSION = "1.0.1";
 
 // src/tools/models.ts
 init_dist();
@@ -32618,6 +32618,8 @@ import { fileURLToPath } from "node:url";
 
 // src/models.ts
 var DEFAULT_IMAGE_MODEL = "gemini-3.1-flash-image";
+var DEFAULT_VIDEO_MODEL = "gemini-omni-flash-preview";
+var DEFAULT_MUSIC_MODEL = "lyria-3-clip-preview";
 function resolveModel(perCall, envOverride) {
   return perCall?.trim() || envOverride?.trim() || DEFAULT_IMAGE_MODEL;
 }
@@ -32881,14 +32883,66 @@ var GeminiClient = class {
       if (opts.searchTypes?.length) searchTool.search_types = opts.searchTypes;
       body.tools = [searchTool];
     }
+    const data = await this.postInteraction(body, opts);
+    const { images, text, grounding } = this.extractInteraction(data, opts.searchTypes);
+    if (images.length === 0) {
+      throw new McpToolError("Gemini returned no image", {
+        hint: "The request may have been blocked by safety filters \u2014 try rephrasing the prompt."
+      });
+    }
+    return { id: data.id, images, text, grounding };
+  }
+  /**
+   * Generate video via the omni model (Interactions API, inline delivery). Same
+   * plumbing as {@link interact} — text/image input parts, a `video`
+   * response_format, optional `video_config.task`, and `previous_interaction_id`
+   * for edits. Preview model: shapes are docs-derived; parsing stays tolerant.
+   */
+  async generateVideo(opts) {
+    const model = opts.model ?? DEFAULT_VIDEO_MODEL;
+    const inputParts = [{ type: "text", text: opts.input }];
+    for (const img of opts.images ?? []) inputParts.push({ type: "image", mime_type: img.mimeType, data: img.base64 });
+    const responseFormat = { type: "video", delivery: "inline" };
+    if (opts.aspectRatio) responseFormat.aspect_ratio = opts.aspectRatio;
+    const body = { model, input: inputParts, response_format: responseFormat };
+    if (opts.task) body.generation_config = { video_config: { task: opts.task } };
+    if (opts.previousInteractionId !== void 0) body.previous_interaction_id = opts.previousInteractionId;
+    const data = await this.postInteraction(body, opts);
+    const { videos, text } = this.extractInteraction(data);
+    if (videos.length === 0) {
+      throw new McpToolError("Gemini returned no video", {
+        hint: "The request may have been blocked by a safety filter, or the clip exceeded the ~4MB inline limit \u2014 try a shorter/simpler prompt or a different aspect ratio."
+      });
+    }
+    return { id: data.id, videos, text };
+  }
+  /**
+   * Generate music via a Lyria model (Interactions API, inline audio). Reuses
+   * the interact plumbing with an `audio` response_format. Preview model.
+   */
+  async generateMusic(opts) {
+    const model = opts.model ?? DEFAULT_MUSIC_MODEL;
+    const inputParts = [{ type: "text", text: opts.input }];
+    for (const img of opts.images ?? []) inputParts.push({ type: "image", mime_type: img.mimeType, data: img.base64 });
+    const responseFormat = { type: "audio" };
+    if (opts.audioFormat) responseFormat.audio_format = opts.audioFormat;
+    const body = { model, input: inputParts, response_format: responseFormat };
+    if (opts.previousInteractionId !== void 0) body.previous_interaction_id = opts.previousInteractionId;
+    const data = await this.postInteraction(body, opts);
+    const { audios, text } = this.extractInteraction(data);
+    if (audios.length === 0) {
+      throw new McpToolError("Gemini returned no audio", {
+        hint: "The request may have been blocked by a safety filter \u2014 try rephrasing the prompt."
+      });
+    }
+    return { id: data.id, audios, text };
+  }
+  /** POST `/interactions` with the chained-404 retry. Shared by all media. */
+  async postInteraction(body, opts) {
     const { interactionsApi } = this.apisFor(resolveTimeoutMs(opts.timeoutMs, opts.imageSize));
-    let data;
     for (let attempt = 0; ; attempt++) {
       try {
-        data = await interactionsApi.fetchJson("POST", "/interactions", {
-          body
-        });
-        break;
+        return await interactionsApi.fetchJson("POST", "/interactions", { body });
       } catch (err) {
         if (!(opts.previousInteractionId && err instanceof ApiError && err.status === 404)) throw err;
         if (attempt < CHAIN_404_RETRIES) {
@@ -32898,13 +32952,23 @@ var GeminiClient = class {
         throw new McpToolError(
           `Previous interaction "${opts.previousInteractionId}" was not found (retried ${CHAIN_404_RETRIES}\xD7 \u2014 the interactions store can lag briefly after a turn completes). If the id is old: interactions are retained 55 days on the paid tier (1 day on the free tier) and are scoped to the API key that created them.`,
           {
-            hint: "Retry once more in a few seconds; if it keeps failing, start a new chain: call gemini_interact again with the last output image passed via `images` (or `images_base64`) plus the same instruction, then chain the NEW interaction id.",
+            hint: "Retry once more in a few seconds; if it keeps failing, start a new chain by re-attaching the last output as an input plus the same instruction, then chain the NEW interaction id.",
             cause: err
           }
         );
       }
     }
+  }
+  /**
+   * Collect `model_output` media (image/video/audio) + text + grounding from an
+   * interactions response. Only the `model_output` step is surfaced — `thought`
+   * steps (internal reasoning / draft images) are dropped on purpose so they
+   * can't leak into text or pollute the outputs. Tolerant of snake/camel MIME.
+   */
+  extractInteraction(data, searchTypes) {
     const images = [];
+    const videos = [];
+    const audios = [];
     const textParts = [];
     const searchQueries = [];
     const searchSuggestions = /* @__PURE__ */ new Set();
@@ -32914,7 +32978,7 @@ var GeminiClient = class {
         continue;
       }
       if (step.type === "google_search_result") {
-        if (opts.searchTypes?.includes("image_search")) {
+        if (searchTypes?.includes("image_search")) {
           for (const r of step.result ?? []) if (r.search_suggestions?.trim()) searchSuggestions.add(r.search_suggestions);
         }
         continue;
@@ -32922,27 +32986,24 @@ var GeminiClient = class {
       if (step.type !== "model_output") continue;
       for (const parts of [step.content ?? [], step.summary ?? []]) {
         for (const part of parts) {
-          if (part.type === "image" && part.data) {
-            images.push({ base64: part.data, mimeType: part.mime_type ?? "image/jpeg" });
+          const mime = part.mime_type ?? part.mimeType;
+          if (part.data) {
+            if (part.type === "image") images.push({ base64: part.data, mimeType: mime ?? "image/jpeg" });
+            else if (part.type === "video") videos.push({ base64: part.data, mimeType: mime ?? "video/mp4" });
+            else if (part.type === "audio") audios.push({ base64: part.data, mimeType: mime ?? "audio/mpeg" });
           } else if (part.type === "text" && part.text?.trim()) {
             textParts.push(part.text);
           }
         }
       }
     }
-    if (images.length === 0) {
-      throw new McpToolError("Gemini returned no image", {
-        hint: "The request may have been blocked by safety filters \u2014 try rephrasing the prompt."
-      });
-    }
-    const resultText = textParts.join("\n") || void 0;
     let grounding;
     if (searchQueries.length > 0 || searchSuggestions.size > 0) {
       grounding = {};
       if (searchQueries.length > 0) grounding.queries = searchQueries;
       if (searchSuggestions.size > 0) grounding.search_suggestions = [...searchSuggestions];
     }
-    return { id: data.id, images, text: resultText, grounding };
+    return { images, videos, audios, text: textParts.join("\n") || void 0, grounding };
   }
 };
 var client = new GeminiClient();
@@ -33004,10 +33065,22 @@ function sniffMimeBytes(buf) {
   }
   return "image/png";
 }
-async function writeImage(dir, base, base643, mimeType) {
+function mediaExt(mimeType) {
+  const m = mimeType.toLowerCase();
+  if (m.includes("jpeg")) return "jpg";
+  if (m.includes("webp")) return "webp";
+  if (m.includes("png")) return "png";
+  if (m.includes("mp4")) return "mp4";
+  if (m.includes("quicktime")) return "mov";
+  if (m.includes("webm")) return "webm";
+  if (m.includes("wav")) return "wav";
+  if (m.includes("mpeg") || m.includes("mp3")) return "mp3";
+  if (m.includes("ogg")) return "ogg";
+  return m.split("/")[1]?.split(";")[0]?.replace(/^x-/, "") || "bin";
+}
+async function writeMedia(dir, base, base643, mimeType) {
   await mkdir(dir, { recursive: true });
-  const ext = mimeType.includes("jpeg") ? "jpg" : mimeType.includes("webp") ? "webp" : "png";
-  const path = await uniquePath(dir, base, ext);
+  const path = await uniquePath(dir, base, mediaExt(mimeType));
   await writeFile(path, Buffer.from(base643, "base64"));
   return resolve(path);
 }
@@ -33104,7 +33177,7 @@ async function gatherImageInputs(opts) {
   return [...clipboardImages, ...fileImages];
 }
 function baseName(name) {
-  const stripped = name.replace(/\.(png|jpe?g|webp)$/i, "");
+  const stripped = name.replace(/\.(png|jpe?g|webp|mp4|mov|webm|mp3|wav|ogg)$/i, "");
   const safe = stripped.replace(/[^A-Za-z0-9._-]+/g, "-").replace(/-+/g, "-").replace(/^-+|-+$/g, "");
   return safe || "image";
 }
@@ -33214,21 +33287,24 @@ function buildMeta(model, seed, opts) {
   if (opts.image_size) meta3.image_size = opts.image_size;
   return meta3;
 }
-async function emit(named, opts, meta3, onWritten) {
-  if (opts.inline) {
-    const imageBlocks = named.map((n) => ({ type: "image", data: n.image.base64, mimeType: n.image.mimeType }));
+async function emitMedia(named, kind, opts, meta3, onWritten) {
+  const inline = opts.inline && kind !== "video";
+  if (inline) {
+    const blocks = named.map((n) => ({ type: kind, data: n.media.base64, mimeType: n.media.mimeType }));
     if (meta3 && Object.keys(meta3).length > 0) {
-      return {
-        content: [{ type: "text", text: JSON.stringify(meta3, null, 2) }, ...imageBlocks]
-      };
+      return { content: [{ type: "text", text: JSON.stringify(meta3, null, 2) }, ...blocks] };
     }
-    return { content: imageBlocks };
+    return { content: blocks };
   }
   const dir = resolveOutputDir(opts.output_dir);
-  const images = [];
-  for (const n of named) images.push(await writeImage(dir, n.base, n.image.base64, n.image.mimeType));
-  await onWritten?.(images);
-  return textResult({ images, ...meta3 });
+  const paths = [];
+  for (const n of named) paths.push(await writeMedia(dir, n.base, n.media.base64, n.media.mimeType));
+  await onWritten?.(paths);
+  const note = opts.inline && kind === "video" ? { inline_unsupported: "video has no MCP inline content type \u2014 written to disk instead" } : {};
+  return textResult({ [`${kind}s`]: paths, ...note, ...meta3 });
+}
+async function emit(named, opts, meta3, onWritten) {
+  return emitMedia(named.map((n) => ({ media: n.image, base: n.base })), "image", opts, meta3, onWritten);
 }
 
 // src/jobs.ts
@@ -33371,7 +33447,7 @@ var SEND_LOCAL_ACTION = "Send local file input(s) to the Gemini API";
 var REFINE_HINT = "To iteratively refine this image, use gemini_interact (pass the output file via `images` on the first call, then chain `previous_interaction_id` on each subsequent call).";
 function registerGenerateTools(server) {
   server.registerTool(
-    "gemini_generate_image",
+    "gemini_image_generate",
     {
       description: "Generate image(s) from a text prompt with a Gemini image model (Nano Banana / Nano Banana Pro). If the result will likely be refined iteratively, prefer gemini_interact (multi-turn) as the entry point.",
       annotations: { readOnlyHint: false, openWorldHint: true },
@@ -33392,7 +33468,7 @@ function registerGenerateTools(server) {
       if (gate) return gate;
       const count = args.count ?? 1;
       const model = resolveModel(args.model, readEnvVar("GEMINI_IMAGE_MODEL"));
-      const fingerprint = fingerprintRequest("gemini_generate_image", {
+      const fingerprint = fingerprintRequest("gemini_image_generate", {
         model,
         prompt: args.prompt,
         aspect_ratio: args.aspect_ratio,
@@ -33406,7 +33482,7 @@ function registerGenerateTools(server) {
         video_url: args.video_url,
         video_path: args.video_path
       });
-      return dispatch({ toolName: "gemini_generate_image", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
+      return dispatch({ toolName: "gemini_image_generate", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
         const seed = pickSeed(args.seed);
         const slug = args.filename ? baseName(args.filename) : slugify2(args.prompt);
         const refInputs = await gatherImageInputs({ images: args.images, images_base64: args.images_base64, from_clipboard: args.from_clipboard });
@@ -33447,9 +33523,9 @@ function registerGenerateTools(server) {
     }
   );
   server.registerTool(
-    "gemini_edit_image",
+    "gemini_image_edit",
     {
-      description: "Edit or compose images: provide one or more input images (paths or base64), plus a text instruction. For a SERIES of successive edits to the same image, prefer gemini_interact (multi-turn) \u2014 it keeps edit context and avoids re-processing the full image each round; use gemini_edit_image for one-off edits or composing multiple distinct inputs. Gemini over-preserves the input; there is no edit-strength control \u2014 for large structural changes, reroll with a different `seed` or more forceful wording.",
+      description: "Edit or compose images: provide one or more input images (paths or base64), plus a text instruction. For a SERIES of successive edits to the same image, prefer gemini_interact (multi-turn) \u2014 it keeps edit context and avoids re-processing the full image each round; use gemini_image_edit for one-off edits or composing multiple distinct inputs. Gemini over-preserves the input; there is no edit-strength control \u2014 for large structural changes, reroll with a different `seed` or more forceful wording.",
       annotations: { readOnlyHint: false, openWorldHint: true },
       inputSchema: {
         prompt: external_exports.string().min(1).describe("Instruction describing the edit or composition"),
@@ -33469,7 +33545,7 @@ function registerGenerateTools(server) {
       const gate = await previewLocalInputsUnlessConfirmed(args.confirm, SEND_LOCAL_ACTION, GENERATE_ENDPOINT, args.images);
       if (gate) return gate;
       const model = resolveModel(args.model, readEnvVar("GEMINI_IMAGE_MODEL"));
-      const fingerprint = fingerprintRequest("gemini_edit_image", {
+      const fingerprint = fingerprintRequest("gemini_image_edit", {
         model,
         prompt: args.prompt,
         aspect_ratio: args.aspect_ratio,
@@ -33480,7 +33556,7 @@ function registerGenerateTools(server) {
         images_base64: args.images_base64,
         from_clipboard: args.from_clipboard
       });
-      return dispatch({ toolName: "gemini_edit_image", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
+      return dispatch({ toolName: "gemini_image_edit", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
         const seed = pickSeed(args.seed);
         const slug = args.filename ? baseName(args.filename) : slugify2(args.prompt);
         const inputs = await gatherImageInputs({ images: args.images, images_base64: args.images_base64, from_clipboard: args.from_clipboard });
@@ -33513,7 +33589,7 @@ init_zod();
 init_dist();
 function registerSetTools(server) {
   server.registerTool(
-    "gemini_generate_set",
+    "gemini_image_set",
     {
       description: "Generate a consistent SET of images: a master image from master_prompt, then one image per scene that references the master so the subject/style stays consistent. Provide `scenes` (explicit per-image prompts) OR `count` (variations of the master).",
       annotations: { readOnlyHint: false, openWorldHint: true },
@@ -33536,7 +33612,7 @@ function registerSetTools(server) {
       const gate = await previewLocalInputsUnlessConfirmed(args.confirm, "Send local master image input(s) to the Gemini API", "/v1beta/models/{model}:generateContent", args.master_images);
       if (gate) return gate;
       const model = resolveModel(args.model, readEnvVar("GEMINI_IMAGE_MODEL"));
-      const fingerprint = fingerprintRequest("gemini_generate_set", {
+      const fingerprint = fingerprintRequest("gemini_image_set", {
         model,
         master_prompt: args.master_prompt,
         scenes: args.scenes,
@@ -33550,7 +33626,7 @@ function registerSetTools(server) {
         master_images_base64: args.master_images_base64,
         from_clipboard: args.from_clipboard
       });
-      return dispatch({ toolName: "gemini_generate_set", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
+      return dispatch({ toolName: "gemini_image_set", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
         const seed = pickSeed(args.seed);
         const cfg = { model: args.model, aspectRatio: args.aspect_ratio, imageSize: args.image_size, thinkingLevel: args.thinking_level, googleSearch: args.google_search, timeoutMs: args.timeout_ms };
         const slug = args.basename ? baseName(args.basename) : slugify2(args.master_prompt);
@@ -33723,6 +33799,172 @@ function registerInteractTools(server) {
   );
 }
 
+// src/tools/video.ts
+init_zod();
+init_dist();
+var VIDEO_ASPECT_RATIOS = ["16:9", "9:16"];
+var VIDEO_TASKS = ["text_to_video", "image_to_video", "reference_to_video", "edit"];
+var lastVideoInteractionId;
+function registerVideoTools(server) {
+  server.registerTool(
+    "gemini_video_generate",
+    {
+      description: 'Generate a short video via the Gemini omni model (preview): text\u2192video, image\u2192video / reference\u2192video (supply reference image[s]), or edit a prior video (task: "edit" + previous_interaction_id / continue_last). Output is written to disk as MP4 (video has no inline MCP block). Video runs long \u2014 use `async: true` to get a job_id immediately and poll gemini_get_result, or raise `timeout_ms`. Preview model: needs a funded account.',
+      annotations: { readOnlyHint: false, openWorldHint: true },
+      inputSchema: {
+        prompt: external_exports.string().min(1).describe("Description of the video to generate (or the edit instruction when task=edit)"),
+        aspect_ratio: external_exports.enum(VIDEO_ASPECT_RATIOS).optional().describe("Output aspect ratio (omni: 16:9 or 9:16)"),
+        task: external_exports.enum(VIDEO_TASKS).optional().describe("text_to_video (default), image_to_video / reference_to_video (need image input), or edit (needs previous_interaction_id)"),
+        images: external_exports.array(external_exports.string().min(1)).optional().describe("Reference image path(s) for image_to_video / reference_to_video"),
+        images_base64: external_exports.array(external_exports.string().min(1)).optional().describe("Reference images as base64 strings or data URIs"),
+        from_clipboard: external_exports.boolean().optional().describe("Use the image currently on the macOS clipboard as a reference"),
+        filename: external_exports.string().optional().describe("Base filename for the output video (extension stripped; default: slugified prompt)"),
+        output_dir: external_exports.string().optional().describe("Directory to write the video to (default: $GEMINI_OUTPUT_DIR or cwd)"),
+        model: external_exports.string().regex(/^[\w.-]+$/, 'must be a bare model id (letters, digits, ".", "_", "-")').optional().describe(`Model id override (default: ${DEFAULT_VIDEO_MODEL})`),
+        previous_interaction_id: external_exports.string().optional().describe('Interaction id to edit/continue (with task: "edit")'),
+        continue_last: external_exports.boolean().optional().describe("Continue from the most recent video interaction this server created (explicit previous_interaction_id wins)"),
+        timeout_ms: timeoutMsSchema,
+        idempotency_key: idempotencyKeySchema,
+        async: asyncSchema,
+        confirm: schemaConfirm
+      }
+    },
+    async (args, extra) => {
+      let previousInteractionId = args.previous_interaction_id;
+      if (!previousInteractionId && args.continue_last) {
+        if (!lastVideoInteractionId) {
+          throw new McpToolError("No previous video interaction to continue in this session.", {
+            hint: "Generate a video first, or pass an explicit previous_interaction_id."
+          });
+        }
+        previousInteractionId = lastVideoInteractionId;
+      }
+      const gate = await previewLocalInputsUnlessConfirmed(args.confirm, "Send local image input(s) to the Gemini omni API", "/v1beta/interactions", args.images);
+      if (gate) return gate;
+      const model = args.model?.trim() || DEFAULT_VIDEO_MODEL;
+      const fingerprint = fingerprintRequest("gemini_video_generate", {
+        model,
+        prompt: args.prompt,
+        aspect_ratio: args.aspect_ratio,
+        task: args.task,
+        images: args.images,
+        images_base64: args.images_base64,
+        from_clipboard: args.from_clipboard,
+        previous_interaction_id: previousInteractionId
+      });
+      return dispatch({ toolName: "gemini_video_generate", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
+        const inputs = await gatherImageInputs({ images: args.images, images_base64: args.images_base64, from_clipboard: args.from_clipboard });
+        const r = await withProgressHeartbeat(extra, `Generating video (${model})`, () => client.generateVideo({
+          input: args.prompt,
+          images: inputs.length ? inputs : void 0,
+          model: args.model,
+          aspectRatio: args.aspect_ratio,
+          task: args.task,
+          previousInteractionId,
+          timeoutMs: args.timeout_ms
+        }));
+        lastVideoInteractionId = r.id;
+        const meta3 = { model, interaction_id: r.id };
+        if (previousInteractionId) meta3.previous_interaction_id = previousInteractionId;
+        if (r.text) meta3.text = r.text;
+        meta3.hint = `To edit this video, call gemini_video_generate again with task: "edit" and previous_interaction_id: "${r.id}" (or continue_last: true).`;
+        const slug = args.filename ? baseName(args.filename) : slugify2(args.prompt);
+        const named = r.videos.map((media, i) => ({
+          media,
+          base: r.videos.length > 1 ? `${slug}-${String(i + 1).padStart(2, "0")}` : slug
+        }));
+        return emitMedia(named, "video", { output_dir: args.output_dir }, meta3);
+      });
+    }
+  );
+}
+
+// src/tools/music.ts
+init_zod();
+init_dist();
+var MUSIC_MODELS = ["lyria-3-clip-preview", "lyria-3-pro-preview"];
+var AUDIO_FORMATS = ["mp3", "wav"];
+function isProModel(model) {
+  return /pro/i.test(model);
+}
+var lastMusicInteractionId;
+function registerMusicTools(server) {
+  server.registerTool(
+    "gemini_music_generate",
+    {
+      description: "Generate music from a text prompt (mood, genre, instruments, structure, or lyrics inline) via a Lyria model (preview): lyria-3-clip-preview (~30s clips, default) or lyria-3-pro-preview (longer, WAV-capable). Written to disk as MP3/WAV (or returned inline). Runs long \u2014 use `async: true` + gemini_get_result, or raise `timeout_ms`. Preview model: needs a funded account.",
+      annotations: { readOnlyHint: false, openWorldHint: true },
+      inputSchema: {
+        prompt: external_exports.string().min(1).describe("Description of the music: mood, genre, instruments, tempo, structure, or lyrics"),
+        model: external_exports.enum(MUSIC_MODELS).optional().describe(`Lyria model (default: ${DEFAULT_MUSIC_MODEL}). Pro is longer-form and supports WAV.`),
+        audio_format: external_exports.enum(AUDIO_FORMATS).optional().describe("Output format (default mp3). wav is lyria-3-pro-preview-only."),
+        images: external_exports.array(external_exports.string().min(1)).optional().describe("Optional reference image path(s) to condition the music"),
+        images_base64: external_exports.array(external_exports.string().min(1)).optional().describe("Reference images as base64 strings or data URIs"),
+        from_clipboard: external_exports.boolean().optional().describe("Use the image currently on the macOS clipboard as a reference"),
+        filename: external_exports.string().optional().describe("Base filename for the output audio (extension stripped; default: slugified prompt)"),
+        output_dir: external_exports.string().optional().describe("Directory to write audio to (default: $GEMINI_OUTPUT_DIR or cwd)"),
+        inline: external_exports.boolean().optional().describe("Return base64 audio inline instead of writing to disk"),
+        previous_interaction_id: external_exports.string().optional().describe("Interaction id to continue from"),
+        continue_last: external_exports.boolean().optional().describe("Continue from the most recent music interaction this server created (explicit previous_interaction_id wins)"),
+        timeout_ms: timeoutMsSchema,
+        idempotency_key: idempotencyKeySchema,
+        async: asyncSchema,
+        confirm: schemaConfirm
+      }
+    },
+    async (args, extra) => {
+      const model = args.model ?? DEFAULT_MUSIC_MODEL;
+      if (args.audio_format === "wav" && !isProModel(model)) {
+        throw new McpToolError(`WAV output requires a Pro model; ${model} only produces MP3.`, {
+          hint: 'Use model: "lyria-3-pro-preview" for WAV, or drop audio_format to get MP3.'
+        });
+      }
+      let previousInteractionId = args.previous_interaction_id;
+      if (!previousInteractionId && args.continue_last) {
+        if (!lastMusicInteractionId) {
+          throw new McpToolError("No previous music interaction to continue in this session.", {
+            hint: "Generate music first, or pass an explicit previous_interaction_id."
+          });
+        }
+        previousInteractionId = lastMusicInteractionId;
+      }
+      const gate = await previewLocalInputsUnlessConfirmed(args.confirm, "Send local image input(s) to the Gemini Lyria API", "/v1beta/interactions", args.images);
+      if (gate) return gate;
+      const fingerprint = fingerprintRequest("gemini_music_generate", {
+        model,
+        prompt: args.prompt,
+        audio_format: args.audio_format,
+        images: args.images,
+        images_base64: args.images_base64,
+        from_clipboard: args.from_clipboard,
+        previous_interaction_id: previousInteractionId
+      });
+      return dispatch({ toolName: "gemini_music_generate", fingerprint, idempotencyKey: args.idempotency_key, async: args.async }, async () => {
+        const inputs = await gatherImageInputs({ images: args.images, images_base64: args.images_base64, from_clipboard: args.from_clipboard });
+        const r = await withProgressHeartbeat(extra, `Generating music (${model})`, () => client.generateMusic({
+          input: args.prompt,
+          images: inputs.length ? inputs : void 0,
+          model: args.model,
+          audioFormat: args.audio_format,
+          previousInteractionId,
+          timeoutMs: args.timeout_ms
+        }));
+        lastMusicInteractionId = r.id;
+        const meta3 = { model, interaction_id: r.id };
+        if (previousInteractionId) meta3.previous_interaction_id = previousInteractionId;
+        if (r.text) meta3.text = r.text;
+        meta3.hint = `To continue this track, call gemini_music_generate again with previous_interaction_id: "${r.id}" (or continue_last: true).`;
+        const slug = args.filename ? baseName(args.filename) : slugify2(args.prompt);
+        const named = r.audios.map((media, i) => ({
+          media,
+          base: r.audios.length > 1 ? `${slug}-${String(i + 1).padStart(2, "0")}` : slug
+        }));
+        return emitMedia(named, "audio", { inline: args.inline, output_dir: args.output_dir }, meta3);
+      });
+    }
+  );
+}
+
 // src/tools/jobs.ts
 init_zod();
 function registerJobTools(server) {
@@ -33744,5 +33986,5 @@ await runMcp({
   name: "gemini-mcp",
   version: VERSION,
   banner: "[gemini-mcp] This project was developed and is maintained by AI (Claude). Use at your own discretion.",
-  tools: [registerModelTools, registerGenerateTools, registerSetTools, registerInteractTools, registerJobTools]
+  tools: [registerModelTools, registerGenerateTools, registerSetTools, registerInteractTools, registerVideoTools, registerMusicTools, registerJobTools]
 });